CN104899280B - The asynchronous image search method of fuzzy correlation based on color histogram and NSCT - Google Patents
The asynchronous image search method of fuzzy correlation based on color histogram and NSCT Download PDFInfo
- Publication number
- CN104899280B CN104899280B CN201510291031.1A CN201510291031A CN104899280B CN 104899280 B CN104899280 B CN 104899280B CN 201510291031 A CN201510291031 A CN 201510291031A CN 104899280 B CN104899280 B CN 104899280B
- Authority
- CN
- China
- Prior art keywords
- mrow
- image
- msub
- color
- msup
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5862—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/40—Analysis of texture
- G06T7/41—Analysis of texture based on statistical description of texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Abstract
The present invention relates to image search method,The color characteristic of image is extracted using color histogram,Using the color vector of color histogram and the two features of the height of color post as retrieval foundation,Similarity is calculated using the fuzzy membership functions in fuzzy set theory,α levels fuzzy relation judges similitude,Non-down sampling contourlet transform (NSCT) is introduced to extract the textural characteristics of image simultaneously,Image is decomposed using NSCT conversion,The average and standard variance of sub-band coefficients in extraction different levels multiple directions are characterized vector,Index as image in image library,And calculate the similarity between image using the fuzzy membership functions in fuzzy set theory,Due to its multiple dimensioned property,Multidirectional and translation invariance,Powerful directional information is remained with after decomposition,The textural characteristics of image can be described more fully with,Finally,Above two algorithm is combined,Image is retrieved with comprehensive characteristics.
Description
Technical field
The present invention relates to image search method, it is specially a kind of using color histogram extraction color characteristic and it is non-under adopt
The textural characteristics of sample profile wave convert extraction carry out the image search method of comprehensive characteristics.
Background technology
The content of piece image statement is very abundant, and it contains the feature of many aspects, merely with a kind of feature simultaneously
The full content of image can not be described.In addition, understanding of the people to image is built upon whole features that human eye can identify
On, it is not to be based only on some feature, if so entering only in terms of some to image to the understanding that image is overall
Row description, often cannot comprehensively describe, and image occur large change (amplification, reduce, translation or rotation etc.) when
Preferable retrieval effectiveness can not usually be obtained.
At present, the image search method of single features can not meet the requirement of user, and comprehensive characteristics are retrieved to obtain
It is widely applied.Image is described using two or more features for comprehensive characteristics retrieval, more can be complete than single feature
The content of image is described to face so that the difference of each image is just further obvious, distinguishes information increase, the inspection drawn according to them
Hitch fruit is also more accurate.
The content of the invention
Color and texture are two kinds of features the most frequently used in image retrieval, and the picture number in image library is numerous, and content is more
It is to vary, color characteristic and textural characteristics are all only capable of describing the part attribute of image, and the feature that different images stress is not
It is certain identical, in order to the attribute of more objective comprehensive description image, more preferable retrieval effectiveness is obtained, the purpose of the present invention exists
Comprehensive characteristics image is carried out in the textural characteristics that the color characteristic and NSCT that provide a kind of comprehensive color histogram extraction extract
Search method.
The present invention adopts the following technical scheme that realization:
A kind of asynchronous image search method of fuzzy correlation based on color histogram and NSCT, comprises the following steps:
(1) it is 16 dimensions by the color quantizing of RGB image, to appointing piece image D and image Q to be retrieved in image library, point
Indescribably take color histogram, specific method is as follows:
By transverse axis of the three-dimensional color value (r, g, b) as color histogram, the three-dimensional color value occurs in entire image
Pixel count as the longitudinal axis, produce image D color histogram and image Q color histogram, then utilize color histogram
Figure extracts color characteristic;
When calculating color histogram, color histogram is sorted step by step from high to low by the height of color post, and determines
The level sequence number of each color post, different color histograms are corresponded into the color post of sequence number as same one-level feature, and carry out phase
Like property measurement;
Because rgb space color is three-dimensional, it assumes that a certain level of correspondence of image D and image Q color histogram
A pair of color vectors are respectively:ci(ri,gi,bi) and cj(rj,gj,bj), then its similarity is expressed as with Gauss member function:
Similarity is calculated using fuzzy membership functions, it is as follows:
Formula (5) is obtainedIn substitution formula 9, α is utilized1Level relation fuzzy matching draws retrieval result, wherein,
Threshold value ɑ1Value can be determined according to experimental result.
WhenMore than or equal to threshold value ɑ1When,Value is 1, it is believed that two features are similar, then carry out next
Step;Otherwise, stop, coming back for next image and handled with image to be checked.
(2) similitude judgement, the height h of corresponding color post, are carried out to the height of corresponding level color postiAnd hjIt is similar
Degree is expressed as:
Similarity is calculated using fuzzy membership functions, it is as follows:
Formula (6) is obtainedIn substitution formula 10, α is utilized2Level relation fuzzy matching draws retrieval result, wherein,
Threshold value ɑ2Value can be determined according to experimental result.
WhenMore than or equal to threshold value ɑ2When,Value is 1, it is believed that two features are similar, then carry out next
Step;Otherwise, stop, coming back for next image and handled with image to be checked.
(3), every piece image in image Q to be retrieved and picture library is obtained successively after the processing of step (1) and (2)
The output retrieval result for going out step (1) and (2) is all 1 all images, as new image library;
(4), to appointing piece image P and image Q to be retrieved to carry out NSCT texture feature extractions respectively in new image library,
NSCT texture feature extraction methods are as follows:
RGB image is converted into gray level image, is to gray level image progress decomposition coefficient { 2,3,4 }, sub-band number 4,8,
16 three layers of NSCT conversion, obtains the sub-band coefficients of 28 subbands, calculates the mean μ of each sub-band coefficients respectivelyiAnd standard variance
σi, mean μiWith standard variance σiCalculation formula it is as follows:
Wherein, Ck(i, j) is the coefficient of k-th of NSCT directional subband, and M × N is the size of the subband, μkIt is k-th of direction
The coefficient average value of subband, σkIt is the factor standard variance of k-th of directional subband;The texture feature vector for obtaining each image is
56 dimensions;M and N represents the ranks number of a two field picture;
That is the texture feature vector f=(μ of image P1,σ1,μ2,σ2,...,μ28,σ28),
Image Q texture feature vector f '=(μ '1,σ′1,μ′2,σ′2,...,μ′28,σ′28)。
(5), in new image library appoint piece image P and image Q to be retrieved each obtain 56 dimension textural characteristics to
Amount carries out Gaussian normalization processing respectively, all characteristic values is all normalized in [- 1,1] section, specific method is as follows:
Gaussian normalization is it is assumed that characteristic vector F distribution meets the Gaussian Profile that average is μ, standard variance is σ
Under the conditions of, characteristic vector is normalized using following formula,
In formula (8), the average and standard variance of this set of mean μ and standard variance σ expression characteristic vector F;
The texture feature vector of image P by Gaussian normalization processing
fP=(μ1P,σ1P,μ2P,σ2P,...,μ28P,σ28P), wherein,Etc. successively
F is calculatedP。
The texture feature vector of image Q by Gaussian normalization processing
f′Q=(μ '1Q,σ′1Q,μ′2Q,σ′2Q,...,μ′28Q,σ′28Q);Wherein,
Etc. f is calculated successively′Q。
To appointing piece image P and image Q to be retrieved, the calculation formula of the similarity of two images as follows in image library:
Wherein,WithImage P and image Q to be retrieved is respectively after Gaussian normalization processing respectively in image library
K-th of textural characteristics component value (including mean μkWith standard variance σk, then be altogether in formula 7 56 numerical value plus and).
(6) similarity, is calculated using fuzzy membership functions, it is as follows:
Formula (7) is obtainedIn substitution formula 11, α is utilized3Level relation fuzzy matching draws retrieval result, its
In, threshold value ɑ3Value can be determined according to experimental result.
WhenMore than or equal to threshold value ɑ3When,It is taken as 1, it is believed that image P and image Q feature phases
Seemingly;Otherwise,It is taken as 0, it is believed that image P and image Q features are dissimilar.
(7), by every piece image in image Q to be retrieved and new picture library by the contrast of step (4) to step (6)
Afterwards, all images that retrieval result is 1 are exported, asynchronous integrated retrieval terminates.
Some principles used in the inventive method are described below below.
1st, the principle on color histogram extraction color characteristic is as follows:
The concept of color histogram is the frequency that statistics different colours occur in entire image in some color model
Rate, often it is used to describe the statistical nature of color of image.But color histogram does not have the space considered residing for every kind of color
Position, it only have recorded the number of pixels that various colors occur, that is to say, that can not be described using color histogram in image
Object or object.In the method by transverse axis of the three-dimensional color value (r, g, b) as color histogram, the three-dimensional color value is whole
Then the pixel count occurred in width image extracts color characteristic as the longitudinal axis using color histogram.
2nd, the principle on NSCT texture feature extractions is as follows:
The features such as multiple dimensioned property and multidirectional based on NSCT, using following algorithm.Coloured image is converted into first
Gray level image, then pass through non-down sampling contourlet transform (Non-Subsampled Contourlet Transform, NSCT)
Conversion is decomposed to gray level image, obtains the sub-band coefficients C under different scale, on different directionsk(i, j), each subband are
Number represents the energy of image, by mean μiWith standard variance σiTextural characteristics as image.To gray level image in experimentation
Three layers of NSCT are carried out to decompose.It is { 2,3,4 } to take decomposition coefficient, then each layer directional subband number is respectively 4,8,16;Calculate each subband
Average and the standard variance textural characteristics that convert to obtain as image NSCT, the texture feature vector of each image for 56 (=
(4+8+16) * 2) dimension.Texture feature vector f1=(μ1,σ1,μ2,σ2,...,μ28,σ28), mean μiWith standard variance σiCalculating
Formula is as follows:
Wherein, Ck(i, j) is the coefficient of k-th of NSCT directional subband, and M × N is the size of the subband, μkIt is k-th of direction
The coefficient average value of subband, σkIt is the factor standard variance of k-th of directional subband.
3rd, the principle on the fuzzy correlation of characteristics of image is as follows:
Assuming that set X=R+, Y=R, then fuzzy relation xSy fuzzy membership functions be expressed as with Gauss type function:
The similarity degree of two characteristic vectors can be calculated above, when whether similar the feature for judging entire image is, often
Often need it is concise represent two determination values of "Yes" or "No", at this moment need to extract from the feature set of image a part with
Know the similar feature of feature, and remove dissimilar feature, this process is referred to as de-fuzzy or clear in fuzzy mathematics
Change process.This purpose can be realized using the ɑ levels relation in fuzzy relation, so as to judge that whether similar feature is:
Wherein c ∈ C, C are set of image characteristics;ɑ is threshold value, for judging that whether similar two kinds of features are.Work as μR(C,Ci) be more than
During equal to threshold value ɑ,It is taken as 1, it is believed that two features are similar, otherwise it is assumed that two features are dissimilar.
3.1st, the fuzzy correlation of color histogram
The similarity degree of two color histograms is generally calculated, the similarity degree for seeking to calculate every a pair of colors post enters again
Row statistics.The color histogram of different coloured images is different, but the dominant hue of each width coloured image can be in color histogram
In embodied.When calculating color histogram, color histogram is sorted step by step from high to low by the height of color post, and really
The level sequence number of fixed each color post, different color histograms are corresponded into the color post of sequence number as same one-level feature, and carried out
Similarity measurement.
Here because the dominant hue of different images might not be identical, i.e., different color histograms correspond to the color post of level
Color-values (abscissa) be not necessarily mutually matching, therefore, to color histogram carry out similarity measurement when, it is necessary to first judge
Color histogram corresponds to the abscissa i.e. corresponding relation of color of level color post, then judges the height of corresponding color post again
Similarity degree.
Because rgb space color is three-dimensional, it assumes that a pair of color vectors of a certain level of two color histograms point
It is not:ci(ri,gi,bi) and cj(rj,gj,bj), then its similarity is expressed as with Gauss member function:
The membership function of above formulaThe similarity degree of two color vectors is mapped in [0,1] closed interval.
Similitude judgement, the height h of corresponding color post are carried out to the height of corresponding level color postiAnd hjSimilarity table
It is shown as:
Above formula by the relevance map of color pillar height degree to [0,1] closed interval,Closer 1, two correspondences
The height of post is closer, and when value is 1, highly equal, two color posts are identical.
It is formula (4) using the ɑ level relations in fuzzy relation if carrying out similitude judgement to above-mentioned color characteristic, wherein α
Value is obtained by experiment.
3.2nd, NSCT fuzzy correlation
NSCT conversion, Decomposition order 3 are carried out to image, each layer decomposition coefficient is respectively { 2,3,4 }, calculates each subband
Texture feature vector f=(the μ that average and standard variance convert as image NSCT1,σ1,μ2,σ2,...,μ28,σ28), for figure
As appointing piece image P and image Q to be retrieved, the calculation formula of the similarity of two images as follows in storehouse:
Wherein,WithImage P and image Q to be retrieved k-th of textural characteristics component value be (i.e. respectively in image library
Each k values correspond to a mean μkWith a standard variance σk).By the mould for calculating image to be retrieved and image in image library
Paste membership function obtains its similarity, image in image library is ranked up according to the order of similarity from big to small, functional value
Closer to 1, then image is more similar.
It is formula (4) using the ɑ level relations in fuzzy relation if carrying out similitude judgement to above-mentioned textural characteristics, wherein α
Value is obtained by experiment.
4th, the normalization on characteristic vector
When carrying out the retrieval of comprehensive characteristics, due to consider the feature of two or more numbers, would have to note
The difference anticipated to different characteristic in physical meaning and numerically.And these differences can frequently result in retrieval error, in order to avoid this
The influence of aspect, it is necessary to which characteristic vector is normalized.The normalization of characteristic vector is generally divided into two classes:Feature to
The inside normalization of amount and the outside normalization of characteristic vector.
(1), the inside normalization of characteristic vector is mainly for each component in a certain characteristic vector, by by its normalizing
Change into a certain particular range to cause contribution of each component to final retrieval result identical.
Color of image feature extraction be image color histogram color post color-values and height, they make respectively
It is characterized, the implication each represented is identical, and span change is also little, therefore need not carry out internal normalization.
Image texture characteristic extraction is to convert the obtained mean μ and standard variance σ of each sub-band coefficients by NSCT, by
It is larger in the mean μ and the gap of the standard variance σ orders of magnitude that gained is converted through NSCT, need exist for using the inside of characteristic vector
Normalization, because this feature meets Gaussian Profile, therefore carries out Gaussian normalization with Gaussian normalization formula to it, will be all
Characteristic value is all normalized in [- 1,1] section.Gaussian normalization is the mark it is assumed that characteristic vector F distribution meets that average is μ
Quasi- variance is the normalization that is carried out by following formula to characteristic vector under conditions of σ Gaussian Profile;
(2), the outside normalization of characteristic vector is mainly for multiple characteristic vectors, by being normalized to ensure each feature
The importance of vector, can also stress a certain feature by changing weight.
Because a fuzzy relation S from X to Y is a mappings of the X × Y to [0,1], so based on color histogram
Similarity and similarity based on NSCT are all distributed in [0,1] section, and their physical significance is identical, and span also determines
In [0,1] section, so need not carry out the outside normalization of characteristic vector.
In summary, this method is reasonable in design, and the color characteristic of image is extracted using color histogram, by color histogram
The color vector and the two features of the height of color post of figure utilize the fuzzy membership letter in fuzzy set theory as retrieval foundation
Number calculates similarity, α levels fuzzy relation judges similitude, while introduces non-down sampling contourlet transform (Non-Subsampled
Contourlet Transform, NSCT) textural characteristics of image are extracted, image is decomposed using NSCT conversion, carried
The average and standard variance for taking the sub-band coefficients in different levels multiple directions are characterized vector, the rope as image in image library
Draw, and the similarity between image is calculated using the fuzzy membership functions in fuzzy set theory, due to its multiple dimensioned property, multi-direction
Property and translation invariance, remain with powerful directional information, the textural characteristics of image can be described more fully with, most after decomposition
Afterwards, above two algorithm is combined, image retrieved with comprehensive characteristics.It is right in this method based on comprehensive characteristics
The setting of weights can influence the effect of image retrieval so that color and textural characteristics can have complementary advantages, and improve the retrieval of image
Precision.This comprehensive characteristics method not only has more preferable retrieval precision than the search method of single features, because it is carried in feature
Improvement in terms of taking with similarity measurement so that it is than other comprehensive characteristics methods also advantageously.
Brief description of the drawings
Fig. 1 represents the schematic flow sheet of the inventive method.
Fig. 2 represents Corel image library examples.
Fig. 3 represents image Q to be retrieved.
Fig. 4 represents the width image of return 30 of the color characteristic retrieval using not given threshold.
Fig. 5 represents the width image of return 30 of the NSCT texture feature extractions retrieval using not given threshold.
Fig. 6 represents the image arrived using the asynchronous integrated retrieval of not given threshold.
Fig. 7 represents the image that the color characteristic of given threshold retrieves.
Fig. 8 represents the image that the NSCT texture feature extractions of given threshold retrieve.
Fig. 9 represents the image retrieved using inventive method.
Embodiment
The specific embodiment of the present invention is described in detail below in conjunction with the accompanying drawings.
The asynchronous image search method of a kind of fuzzy correlation based on color histogram and NSCT, as shown in figure 1, including as follows
Step:
(1) it is 16 dimensions by the color quantizing of RGB image, to appointing piece image D and image Q to be retrieved in image library, point
Indescribably take color histogram, specific method is as follows:
By transverse axis of the three-dimensional color value (r, g, b) as color histogram, the three-dimensional color value occurs in entire image
Pixel count as the longitudinal axis, produce image D color histogram and image Q color histogram, then utilize color histogram
Figure extracts color characteristic.
When calculating color histogram, color histogram is sorted step by step from high to low by the height of color post, and determines
The level sequence number of each color post, different color histograms are corresponded into the color post of sequence number as same one-level feature, and carry out phase
Like property measurement.
Because rgb space color is three-dimensional, it assumes that a certain level of correspondence of image D and image Q color histogram is (excellent
Elect the first level, i.e. height highest level as, be advantageous to calculate it is accurate) a pair of color vectors be respectively:ci(ri,gi,
bi) and cj(rj,gj,bj), then its similarity is expressed as with Gauss member function:
The membership function of above formulaThe similarity degree of two color vectors is mapped in [0,1] closed interval.
Similarity is calculated using fuzzy membership functions, it is as follows:
Formula (5) is obtainedIn substitution formula 9, α is utilized1Level relation fuzzy matching draws retrieval result, wherein,
Pass through many experiments threshold value ɑ1Value is 0.95;
WhenMore than or equal to threshold value ɑ1When,Value is 1, it is believed that two features are similar, then carry out next
Step;Otherwise, stop, coming back for next image and handled with image to be checked.
(2) similitude judgement, the height h of corresponding color post, are carried out to the height of corresponding level color postiAnd hjIt is similar
Degree is expressed as:
Above formula by the relevance map of color pillar height degree to [0,1] closed interval,Closer 1, two correspondences
The height of post is closer, and when value is 1, highly equal, two color posts are identical.
Similarity is calculated using fuzzy membership functions, it is as follows:
Formula (6) is obtainedIn substitution formula 10, α is utilized2Level relation fuzzy matching draws retrieval result, wherein,
Pass through many experiments threshold value ɑ2Value is 0.90;
WhenMore than or equal to threshold value ɑ2When,Value is 1, it is believed that two features are similar, then carry out next
Step;Otherwise, stop, coming back for next image and handled with image to be checked.
(3), every piece image in image Q to be retrieved and picture library is obtained successively after the processing of step (1) and (2)
The output retrieval result for going out step (1) and (2) is all 1 all images, as new image library;
(4), to appointing piece image P and image Q to be retrieved to carry out NSCT texture feature extractions respectively in new image library,
NSCT texture feature extraction methods are as follows:
RGB image is converted into gray level image, is to gray level image progress decomposition coefficient { 2,3,4 }, sub-band number 4,8,
16 three layers of NSCT conversion, obtains the sub-band coefficients of 28 subbands, calculates the mean μ of each sub-band coefficients respectivelyiAnd standard variance
σi, mean μiWith standard variance σiCalculation formula it is as follows:
Wherein, Ck(i, j) is the coefficient of k-th of NSCT directional subband, and M × N is the size of the subband, μkIt is k-th of direction
The coefficient average value of subband, σkIt is the factor standard variance of k-th of directional subband;The texture feature vector for obtaining each image is
56 dimensions;M and N represents the ranks number of a two field picture;
That is the texture feature vector f=(μ of image P1,σ1,μ2,σ2,...,μ28,σ28),
Image Q texture feature vector f '=(μ '1,σ′1,μ′2,σ′2,...,μ′28,σ′28)。
(5), in new image library appoint piece image P and image Q to be retrieved each obtain 56 dimension textural characteristics to
Amount carries out Gaussian normalization processing respectively, all characteristic values is all normalized in [- 1,1] section, specific method is as follows:
Gaussian normalization is it is assumed that characteristic vector F distribution meets the Gaussian Profile that average is μ, standard variance is σ
Under the conditions of, characteristic vector is normalized using following formula,
In formula (8), the average and standard variance of this set of mean μ and standard variance σ expression characteristic vector F;
The texture feature vector of image P by Gaussian normalization processing
fP=(μ1P,σ1P,μ2P,σ2P,...,μ28P,σ28P), wherein,Etc. successively
F is calculatedP。
The texture feature vector of image Q by Gaussian normalization processing
f′Q=(μ '1Q,σ′1Q,μ′2Q,σ′2Q..., μ '28Q,σ′28Q), wherein,
Etc. f is calculated successively′Q。
To appointing piece image P and image Q to be retrieved, the calculation formula of the similarity of two images as follows in image library:
Wherein,WithImage P and image Q to be retrieved is respectively after Gaussian normalization processing respectively in image library
K-th of textural characteristics component value (including mean μkWith standard variance σk)。
(6) similarity, is calculated using fuzzy membership functions, it is as follows:
Formula (7) is obtainedIn substitution formula 11, α is utilized3Level relation fuzzy matching draws retrieval result, its
In, pass through many experiments threshold value ɑ3Value is 0.75;
WhenMore than or equal to threshold value ɑ3When,It is taken as 1, it is believed that image P is similar with image Q features;
Otherwise,It is taken as 0, it is believed that image P and image Q features are dissimilar.
(7), by every piece image in image Q to be retrieved and new picture library by the processing of step (4) to step (6)
Afterwards, all images of output retrieval result 1 are drawn, asynchronous integrated retrieval terminates.
The technique effect of the inventive method is analyzed below by specific experimental result.
As shown in Fig. 2 the image library used in experiment is from 10 class cromograms in the Corel picture libraries of Stanford Univ USA
Picture, per class 100 width, totally 1000 width image.
Searching system is " image indexing system based on NSCT&Color fuzzy correlations ", is carried out with above-mentioned image library real
Test, evaluation criterion selection precision ratio, select 5 width images at random from every class image respectively as image to be retrieved, calculate respectively
To the precision ratio of each image, the average retrieval precision ratio to every a kind of image is then calculated.
The contrast experiment that single features and comprehensive characteristics influence on retrieval result
By 1. single color characteristic retrieval mode in experiment;2. single textural characteristics retrieval mode;3. color characteristic
Combined with the asynchronous integrated retrieval of textural characteristics;Three kinds of searching algorithms are compared.
The group tests retrieval result such as table 1.
The searching algorithm performance comparision of table 1 (average precision that image is returned by threshold value)
As it can be seen from table 1 retrieval precision ratio of the simultaneous synthesis searching algorithm to all types image is all higher.Threshold value
Setting cause system to be easier to provide desired result.And the image retrieval precision ratio of asynchronous integrated retrieval algorithm is more up to
100%, this is due to that whole retrieving take into account all two feature vectors, and and if only if, and all characteristic vectors all match
When ability returning result, retrieval precision is higher.
Generally speaking, due to having merged color histogram and NSCT conversion and the advantages of fuzzy set theory three, this paper's
Comprehensive characteristics algorithm is good all than contrast experiment algorithm to the retrieval performance of every a kind of image.
In addition, illustrate the retrieval result based on different characteristic using horse as image to be retrieved with reference to accompanying drawing 3-9.
Fig. 3 is the image using green meadow as small one and large one two brownish red horses of background;It is uncorrelated to have two width in Fig. 4
Elephant image, have the image that 15 width are two brownish red horses in remaining 28 width;It is incoherent image to have six width in Fig. 5, is remained
It is almost small one and large one two dry goods entirely to remaining in 24 width;All it is the two dry goods images using green meadow as background in Fig. 6.
When according to threshold value returning result, what Fig. 7 was retrieved be exactly image to be retrieved in itself;Fig. 8 includes figure to be retrieved
The image of all two dry goods in the 5 width images as including, and first three width is two brownish red horses of green meadow background
Image, preceding two images be even more it is quite similar;Retrieved in Fig. 9 be exactly image to be retrieved in itself.
Contrast understands that the algorithm of comprehensive characteristics is algorithm more effectively a kind of than single features algorithm, improves image retrieval
Accuracy rate.When wanting to return same class image, directly can be returned according to similarity;If go for most like figure
Picture, then can be with given threshold.
The inventive method proposes the fuzzy correlation image inspection of comprehensive color histogram color characteristic and NSCT textural characteristics
Suo Fangfa, compared with single color characteristic retrieval and single textural characteristics retrieval mode, test result indicates that comprehensive
Close characteristic key method and be better than single features search method;And by comprehensive characteristics search method and it is based on profile wave convert and accumulation
The Euclidean distance associated picture search method of color histogram compares experiment, from comparative result, comprehensive characteristics
Algorithm is better than the retrieval effectiveness of the algorithm of single features.
It should be noted last that the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted, although ginseng
It is described in detail according to the embodiment of the present invention, it will be understood by those within the art that, to technical scheme
Modify or equivalent substitution, without departure from the spirit and scope of technical scheme, it all should cover the present invention's
In claims.
Claims (2)
- A kind of 1. asynchronous image search method of fuzzy correlation based on color histogram and NSCT, it is characterised in that:Including as follows Step:(1), to appointing piece image D and image Q to be retrieved in image library, it is 16 dimensions by the color quantizing of RGB image, carries respectively Take color histogram, specific method is as follows:By transverse axis of the three-dimensional color value (r, g, b) as color histogram, the picture that the three-dimensional color value occurs in entire image Prime number produces image D color histogram and image Q color histogram as the longitudinal axis, then using color histogram come Extract color characteristic;When calculating color histogram, color histogram is sorted step by step from high to low by the height of color post, and determines each The level sequence number of color post, different color histograms are corresponded into the color post of sequence number as same one-level feature, and carry out similitude Measurement;Because rgb space color is three-dimensional, then a pair of colors of a certain level of the correspondence of image D and image Q color histogram Vector is respectively:ci(ri,gi,bi) and cj(rj,gj,bj), then its similarity is expressed as with Gauss member function:<mrow> <msub> <mi>&mu;</mi> <mover> <mi>R</mi> <mo>~</mo> </mover> </msub> <mrow> <mo>(</mo> <msub> <mi>c</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>c</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>e</mi> <mrow> <mo>-</mo> <mo>&lsqb;</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>r</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>r</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>g</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>g</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>+</mo> <msup> <mrow> <mo>(</mo> <msub> <mi>b</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>b</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>&rsqb;</mo> </mrow> </msup> <mn>...</mn> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow>Similarity is calculated using fuzzy membership functions, it is as follows:Formula (5) is obtainedIn substitution formula 9, α is utilized1Level relation fuzzy matching draws retrieval result,WhenMore than or equal to threshold value ɑ1When,Value is 1, it is believed that two features are similar, then carry out in next step;It is no Then, stop, coming back for next image and image Q processing to be retrieved;(2) similitude judgement, the height h of corresponding level color post, are carried out to the height of corresponding level color postiAnd hjIt is similar Degree is expressed as:<mrow> <msub> <mi>&mu;</mi> <mover> <mi>S</mi> <mo>~</mo> </mover> </msub> <mrow> <mo>(</mo> <msub> <mi>h</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>h</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <msub> <mi>h</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>h</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>/</mo> <mi>m</mi> <mi>a</mi> <mi>x</mi> <mrow> <mo>(</mo> <msub> <mi>h</mi> <mi>i</mi> </msub> <mo>,</mo> <msub> <mi>h</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mn>...</mn> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow>Similarity is calculated using fuzzy membership functions, it is as follows:Formula (6) is obtainedIn substitution formula 10, α is utilized2Level relation fuzzy matching draws retrieval result,WhenMore than or equal to threshold value ɑ2When,Value is 1, it is believed that two features are similar, then carry out in next step;It is no Then, stop, coming back for next image and image Q processing to be retrieved;(3) every piece image in image Q to be retrieved and picture library, is drawn into step successively after the processing of step (1) and (2) Suddenly the output retrieval result of (1) and (2) is all 1 all images, as new image library;(4), to appointing piece image P and image Q to be retrieved to carry out NSCT texture feature extractions, NSCT respectively in new image library Texture feature extraction method is as follows:RGB image is converted into gray level image, it is { 2,3,4 } to carry out decomposition coefficient to gray level image, sub-band number 4,8,16 Three layers of NSCT conversion, obtain the sub-band coefficients of 28 subbands, calculate the mean μ of each sub-band coefficients respectivelyiWith standard variance σi, Value μiWith standard variance σiCalculation formula it is as follows:<mrow> <msub> <mi>&mu;</mi> <mi>k</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mi>M</mi> <mi>N</mi> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mo>|</mo> <msub> <mi>C</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>|</mo> <mn>...</mn> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow><mrow> <msub> <mi>&sigma;</mi> <mi>k</mi> </msub> <mo>=</mo> <msup> <mrow> <mo>&lsqb;</mo> <mfrac> <mn>1</mn> <mrow> <mi>M</mi> <mi>N</mi> </mrow> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>M</mi> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <msup> <mrow> <mo>(</mo> <mo>|</mo> <mrow> <msub> <mi>C</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> </mrow> <mo>|</mo> <mo>-</mo> <msub> <mi>&mu;</mi> <mi>k</mi> </msub> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>&rsqb;</mo> </mrow> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msup> <mn>...</mn> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow>Wherein, Ck(i, j) is the coefficient of k-th of NSCT directional subband, and M × N is the size of the subband, μkIt is k-th of directional subband Coefficient average value, σkIt is the factor standard variance of k-th of directional subband;The texture feature vector for obtaining each image is 56 Dimension;M and N represents the ranks number of a two field picture;That is the texture feature vector f=(μ of image P1,σ1,μ2,σ2,...,μ28,σ28),Image Q texture feature vector f '=(μ '1,σ′1,μ′2,σ′2,...,μ′28,σ′28);(5), the 56 dimension texture feature vectors for appointing piece image P and image Q to be retrieved each to obtain in new image library are divided Gaussian normalization processing is not carried out, all characteristic values is all normalized in [- 1,1] section, specific method is as follows:Gaussian normalization is under conditions of characteristic vector F distribution meets the Gaussian Profile that average is μ, standard variance is σ, is adopted Characteristic vector is normalized with following formula,<mrow> <msup> <mi>F</mi> <mo>&prime;</mo> </msup> <mo>=</mo> <mfrac> <mrow> <mi>F</mi> <mo>-</mo> <mi>&mu;</mi> </mrow> <mi>&sigma;</mi> </mfrac> <mn>...</mn> <mrow> <mo>(</mo> <mn>8</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow>In formula (8), the average and standard variance of this set of mean μ and standard variance σ expression characteristic vector F;The texture feature vector of image P by Gaussian normalization processingfP=(μ1P,σ1P,μ2P,σ2P,...,μ28P,σ28P),The texture feature vector of image Q by Gaussian normalization processingf′Q=(μ '1Q,σ′1Q,μ′2Q,σ′2Q,...,μ′28Q,σ′28Q);To appointing piece image P and image Q to be retrieved, the calculation formula of the similarity of two images as follows in image library:<mrow> <msub> <mi>&mu;</mi> <mover> <mi>n</mi> <mo>~</mo> </mover> </msub> <mrow> <mo>(</mo> <mi>Q</mi> <mo>,</mo> <mi>P</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>exp</mi> <mo>{</mo> <mo>-</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>28</mn> </munderover> <msup> <mrow> <mo>(</mo> <msubsup> <mi>f</mi> <mi>k</mi> <mrow> <mo>&prime;</mo> <mi>Q</mi> </mrow> </msubsup> <mo>-</mo> <msubsup> <mi>f</mi> <mi>k</mi> <mi>P</mi> </msubsup> <mo>)</mo> </mrow> <mn>2</mn> </msup> <mo>}</mo> <mn>...</mn> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> <mo>,</mo> </mrow>Wherein,WithThe image P and image Q to be retrieved kth after Gaussian normalization processing respectively respectively in image library Individual textural characteristics component value;(6) similarity, is calculated using fuzzy membership functions, it is as follows:Formula (7) is obtainedIn substitution formula 11, α is utilized3Level relation fuzzy matching draws retrieval result,WhenMore than or equal to threshold value ɑ3When,It is taken as 1, it is believed that image P is similar with image Q features;Otherwise,It is taken as 0, it is believed that image P and image Q features are dissimilar;(7), by after processing of the every piece image in image Q to be retrieved and new picture library by step (4) to step (6), obtain Go out all images that output retrieval result is 1 in step (6), asynchronous integrated retrieval terminates.
- 2. the asynchronous image search method of the fuzzy correlation according to claim 1 based on color histogram and NSCT, it is special Sign is:In step (1), threshold value ɑ1For 0.95;In step (2), threshold value ɑ2For 0.90;In step (6), threshold value ɑ3For 0.75.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510291031.1A CN104899280B (en) | 2015-05-29 | 2015-05-29 | The asynchronous image search method of fuzzy correlation based on color histogram and NSCT |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510291031.1A CN104899280B (en) | 2015-05-29 | 2015-05-29 | The asynchronous image search method of fuzzy correlation based on color histogram and NSCT |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104899280A CN104899280A (en) | 2015-09-09 |
CN104899280B true CN104899280B (en) | 2017-12-29 |
Family
ID=54031943
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510291031.1A Expired - Fee Related CN104899280B (en) | 2015-05-29 | 2015-05-29 | The asynchronous image search method of fuzzy correlation based on color histogram and NSCT |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104899280B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CA3034688C (en) | 2017-06-30 | 2021-11-30 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for verifying authenticity of id photo |
CN107358635B (en) * | 2017-07-19 | 2020-11-03 | 辽宁工程技术大学 | Color morphological image processing method based on fuzzy similarity |
CN108052981B (en) * | 2017-12-22 | 2021-08-17 | 辽宁师范大学 | Image classification method based on nonsubsampled Contourlet transformation and convolutional neural network |
CN110321452B (en) * | 2019-05-05 | 2022-08-09 | 广西师范大学 | Image retrieval method based on direction selection mechanism |
CN110694942B (en) * | 2019-10-21 | 2021-05-14 | 武汉纺织大学 | Color difference sorting method for solar cells |
CN111288928B (en) * | 2020-03-12 | 2021-08-13 | 华侨大学 | Object surface three-dimensional topography feature measuring method, device, equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6961736B1 (en) * | 2002-05-31 | 2005-11-01 | Adobe Systems Incorporated | Compact color feature vector representation |
CN102508901A (en) * | 2011-11-07 | 2012-06-20 | 康佳集团股份有限公司 | Content-based massive image search method and content-based massive image search system |
CN104156433A (en) * | 2014-08-11 | 2014-11-19 | 合肥工业大学 | Image retrieval method based on semantic mapping space construction |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100422805B1 (en) * | 2000-12-26 | 2004-03-12 | 한국전자통신연구원 | Color and shape search method and apparatus of image data based on natural language with fuzzy concept |
-
2015
- 2015-05-29 CN CN201510291031.1A patent/CN104899280B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6961736B1 (en) * | 2002-05-31 | 2005-11-01 | Adobe Systems Incorporated | Compact color feature vector representation |
CN102508901A (en) * | 2011-11-07 | 2012-06-20 | 康佳集团股份有限公司 | Content-based massive image search method and content-based massive image search system |
CN104156433A (en) * | 2014-08-11 | 2014-11-19 | 合肥工业大学 | Image retrieval method based on semantic mapping space construction |
Non-Patent Citations (4)
Title |
---|
A fused features approach on Content-Based Image Retrieval based on Fuzzy rule-set;Nikita Raina.etc;《2014 International Conference on Advances in Computing, Communications and Informatics (ICACCI)》;20141201;第2436-2442页 * |
FCTH: Fuzzy Color and Texture Histogram - A Low Level Feature for Accurate Image Retrieval;Savvas A. Chatzichristofis.etc;《2008 Ninth International Workshop on Image Analysis for Multimedia Interactive Services》;20080715;第191-196页 * |
基于多尺度几何分析的纹理图像检索研究;朱正礼;《中国优秀博士学位论文全文数据库》;20130630;I138-32 * |
基于模糊相关的彩色图像检索;梁艳梅等;《中国科学(E辑)》;20031031;第33卷(第10期);第934-938段 * |
Also Published As
Publication number | Publication date |
---|---|
CN104899280A (en) | 2015-09-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104899280B (en) | The asynchronous image search method of fuzzy correlation based on color histogram and NSCT | |
CN104915400B (en) | Fuzzy correlation synchronous images search method based on color histogram and NSCT | |
CN102663391B (en) | Image multifeature extraction and fusion method and system | |
CN103886760B (en) | Real-time vehicle detecting system based on traffic video | |
Loghmani et al. | Recurrent convolutional fusion for RGB-D object recognition | |
CN104063719A (en) | Method and device for pedestrian detection based on depth convolutional network | |
CN103761295B (en) | Automatic picture classification based customized feature extraction method for art pictures | |
CN104134071A (en) | Deformable part model object detection method based on color description | |
CN104240256A (en) | Image salient detecting method based on layering sparse modeling | |
CN110110783A (en) | A kind of deep learning object detection method based on the connection of multilayer feature figure | |
CN103336835B (en) | Image retrieval method based on weight color-sift characteristic dictionary | |
CN110210534B (en) | Multi-packet fusion-based high-resolution remote sensing image scene multi-label classification method | |
CN103218617B (en) | A kind of feature extracting method of polyteny Large space | |
CN106056082A (en) | Video action recognition method based on sparse low-rank coding | |
Duan et al. | Image classification of fashion-MNIST data set based on VGG network | |
CN111881716A (en) | Pedestrian re-identification method based on multi-view-angle generation countermeasure network | |
CN108573241A (en) | A kind of video behavior recognition methods based on fusion feature | |
Deng et al. | Citrus disease recognition based on weighted scalable vocabulary tree | |
CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
CN111709291A (en) | Takeaway personnel identity identification method based on fusion information | |
Huang et al. | A method for object-based color image retrieval | |
Selvarajah et al. | Combined feature descriptor for content based image retrieval | |
Li et al. | CDF space covariance matrix of Gabor wavelet with convolutional neural network for texture recognition | |
Ribas et al. | Fractal dimension of bag-of-visual words | |
Papp et al. | MMKK++ algorithm for clustering heterogeneous images into an unknown number of clusters |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20171229 Termination date: 20200529 |
|
CF01 | Termination of patent right due to non-payment of annual fee |